Despite the fact that Nvidia propelled its 21 billion transistor Volta GPU engineering back in May, up to this point the chip has been utilized only in register cards—particularly, the Tesla V100 cards, which cost about $10,000 for the PCIe adaptation. In any case, now Volta GPU is accessible in a designs card: at the 2017 Neural Information Processing Systems meeting, Nvidia CEO Jen-Hsun Huang declared the Titan V, a $3,000 brilliant video card making the same GPU—the Volta engineering GV100—accessible to general end clients.
Not at all like the Tesla cards, the Titan V is a legitimate designs card. It has three DisplayPort and one HDMI yields, it utilizes the standard GeForce driver stack, and it will play recreations. It’ll most likely play them great, in spite of the fact that Nvidia hasn’t made any gaming execution guarantees hitherto. That is on the grounds that, despite the fact that it’s a designs card, it’s not by any stretch of the imagination being gone for illustrations applications. Or maybe, it’s being gone for similar sorts of GPU-based calculation (particularly machine learning) applications that the Tesla cards are implied for, just on a marginally littler scale. Titan V would be utilized as a part of a workstation PC, as opposed to a process bunch in a data center.
Because of these comparative parts, it’s not very astounding that Titan V’s specs are to a great degree like Tesla V100’s. The two cards have 5,120 figure centers, and both have 640 machines learning-focused “tensor centers” that have some expertise in 4×4 lattice math. Correct clock speeds for the Tesla cards aren’t known, however, Titan V can lift 1.455GHz, somewhat more than V100’s 1.370GHz. It’s the memory subsystem that has the greatest distinction: Titan V has 12GB of HBM2 memory, with a 3,072-piece memory transport. Tesla V100 has 16GB of HBM2 and a 4,096-piece transport. The Titan’s memory is likewise a hair slower, timed at 1.70Gb/s contrasted with 1.75Gb/s.
It truly is a designs card, subsequently its illustrations yields.
Develop/It truly is an illustrations card, thus its designs yields.
Taken together, the new card has practically indistinguishable computational execution—13.8 trillion single exactness gliding point operations every second, 6.9 twofold accuracy TFLOPS, and 110 diminished accuracy machine learning TFLOPS in Titan V, contrasted with 14 single exactness, seven twofold accuracies, and 112 machine learning TFLOPS on V100—yet with memory data transmission of a simple 653GB/s, contrasted with 900GB/s for the registered card.
The high cost and process center imply that regardless of its physical appearance, the Titan V is pointed pretty solidly at the “Ace” side of the “prosumer” client that the Titan line has customarily been gone for. It’s inescapable that some profound took gamers will get a Titan V and utilize it as just a designs card, yet that is surely not the center market.